# Qwen Flash-Next: RTX 5090 CPU offload and long-context gates

The pinned MXFP4 GGUF runs through ik_llama.cpp merge d854dfd36a7262fc46cb7542716b04a9f6d4b47d on one RTX 5090 with WSL2, a 128 GiB-class host and 96 GiB VM ceiling. The baseline puts 38 expert layers on CPU, keeps mmap, disables the CUDA pinned-host override and uses Q8_0 KV. One request and one slot are used throughout; this is not full-model GPU residence.

- Type: Performance report
- Evidence: source_reported; site reproduced: false
- Model family: Qwen
- Model: jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF
- Original author: kutaelee
- Original project: https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe
- Pinned document: https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md
- Source revision: fca75dd108ce04b6a79db3b45934ed58bba893bb
- Source commit time (not experiment date): 2026-09-18T12:12:22Z
- Checked: 2026-10-18
- Experiment date: None
- Runtime topics: llama.cpp
- Hardware topics: nvidia
- Reported setup: {"system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling", "node_count": 1, "parallelism": null, "kv_dtype": "Q8_0", "context_limit_tokens": 32768, "experiment_date": null}
- Performance figures: attributed selections with per-point conditions; no cross-source ranking

## Conditions

Unknown is null. Recipe options are not measured benchmark settings.

```json
{
  "conditions": {
    "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
    "weight_hash": null,
    "runtime_execution_revision": null,
    "input_tokens": null,
    "output_tokens": null,
    "concurrency": null,
    "cache_state": null,
    "quantization": null,
    "speculative_decoding": null,
    "temperature": null,
    "top_p": null
  },
  "test_profiles": [
    {
      "label": "32K needle gate prefill_rate",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L53",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 26253,
        "context_limit_tokens": 32768,
        "cache_state": null,
        "workload": "32K needle gate",
        "memory_cohort": "completed 32K monitoring"
      }
    },
    {
      "label": "32K needle gate decode_rate",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L53",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 26253,
        "context_limit_tokens": 32768,
        "cache_state": null,
        "workload": "32K needle gate",
        "memory_cohort": "completed 32K monitoring"
      }
    },
    {
      "label": "128K server smoke prefill_rate",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L54",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 61,
        "context_limit_tokens": 131072,
        "cache_state": null,
        "workload": "128K server smoke",
        "memory_cohort": "no paired 128K telemetry published"
      }
    },
    {
      "label": "128K server smoke decode_rate",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L54",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 61,
        "context_limit_tokens": 131072,
        "cache_state": null,
        "workload": "128K server smoke",
        "memory_cohort": "no paired 128K telemetry published"
      }
    },
    {
      "label": "128K long-context needle prefill_rate",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L55",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 104805,
        "context_limit_tokens": 131072,
        "cache_state": null,
        "workload": "128K long-context needle",
        "memory_cohort": "no paired 128K telemetry published"
      }
    },
    {
      "label": "128K long-context needle decode_rate",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L55",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 104805,
        "context_limit_tokens": 131072,
        "cache_state": null,
        "workload": "128K long-context needle",
        "memory_cohort": "no paired 128K telemetry published"
      }
    },
    {
      "label": "32K telemetry startup",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L57",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 26253,
        "context_limit_tokens": 32768,
        "cache_state": null,
        "memory_cohort": "completed 32K gate only; not paired with the 104805-token gate",
        "memory_kind": "startup_duration",
        "MemAvailable_sampling_scope": null,
        "VM_ceiling_and_host_MemAvailable_same_scope_verified": false
      }
    },
    {
      "label": "32K telemetry host_rss",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L57",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 26253,
        "context_limit_tokens": 32768,
        "cache_state": null,
        "memory_cohort": "completed 32K gate only; not paired with the 104805-token gate",
        "memory_kind": "peak_host_RSS",
        "MemAvailable_sampling_scope": null,
        "VM_ceiling_and_host_MemAvailable_same_scope_verified": false
      }
    },
    {
      "label": "32K telemetry minimum_memavailable",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L57",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 26253,
        "context_limit_tokens": 32768,
        "cache_state": null,
        "memory_cohort": "completed 32K gate only; not paired with the 104805-token gate",
        "memory_kind": "minimum_MemAvailable_in_reported_monitor_scope",
        "MemAvailable_sampling_scope": null,
        "VM_ceiling_and_host_MemAvailable_same_scope_verified": false
      }
    },
    {
      "label": "32K telemetry swap_use",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L57",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 26253,
        "context_limit_tokens": 32768,
        "cache_state": null,
        "memory_cohort": "completed 32K gate only; not paired with the 104805-token gate",
        "memory_kind": "maximum_swap_use",
        "MemAvailable_sampling_scope": null,
        "VM_ceiling_and_host_MemAvailable_same_scope_verified": false
      }
    },
    {
      "label": "32K telemetry GPU_memory",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L57",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 0,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": 26253,
        "context_limit_tokens": 32768,
        "cache_state": null,
        "memory_cohort": "completed 32K gate only; not paired with the 104805-token gate",
        "memory_kind": "peak_GPU_memory",
        "MemAvailable_sampling_scope": null,
        "VM_ceiling_and_host_MemAvailable_same_scope_verified": false
      }
    },
    {
      "label": "canonical MTP n1 peak GPU memory",
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L61",
      "conditions": {
        "model": "Qwen3.8-Flash-Next",
        "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
        "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
        "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
        "engine": "ik_llama.cpp",
        "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
        "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
        "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
        "node_count": 1,
        "concurrency": 1,
        "configured_max_requests": 1,
        "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
        "CPU_expert_layers": 38,
        "defer_experts": true,
        "mmap": true,
        "GGML_CUDA_NO_PINNED": true,
        "kv_dtype": "Q8_0",
        "draft_tokens": 1,
        "ngram_speculation": false,
        "ngram_embedding_table_present": true,
        "temperature": null,
        "top_p": null,
        "thinking": null,
        "output_tokens": null,
        "actual_output_tokens": null,
        "repetitions": null,
        "experiment_date": null,
        "throughput_origin": "server timing fields, not a client estimate",
        "input_tokens": null,
        "context_limit_tokens": 131072,
        "cache_state": null,
        "memory_cohort": "canonical MTP n1 short gates; not baseline or long-context peak",
        "memory_kind": "peak_GPU_memory"
      }
    }
  ],
  "recipe_parameters": null,
  "deployment_options": []
}
```

## Reading notes

- One source document can contain different experiments and versions; not an independent experiment count.

## Limits

- No independent site model execution.
- Missing sampling, exact executed image digests and measurement boundaries remain null.
- Configured window, completed input, scheduler ceiling and measured full-window concurrency are separate quantities.

## License declaration

{"declared_spdx": "mit", "scope": "Source declaration; models, runtime and dependencies retain their exact terms."}


## Editorial answer

The pinned MXFP4 GGUF runs through ik_llama.cpp merge d854dfd36a7262fc46cb7542716b04a9f6d4b47d on one RTX 5090 with WSL2, a 128 GiB-class host and 96 GiB VM ceiling. The baseline puts 38 expert layers on CPU, keeps mmap, disables the CUDA pinned-host override and uses Q8_0 KV. One request and one slot are used throughout; this is not full-model GPU residence.

By Run AI Here editorial · 2026-10-18

Original source-based interpretation; local Spark Qwen draft with Codex factual review, no site model benchmark.

## Which memory placement belongs to the report?

The pinned MXFP4 GGUF runs through ik_llama.cpp merge d854dfd36a7262fc46cb7542716b04a9f6d4b47d on one RTX 5090 with WSL2, a 128 GiB-class host and 96 GiB VM ceiling. The baseline puts 38 expert layers on CPU, keeps mmap, disables the CUDA pinned-host override and uses Q8_0 KV. One request and one slot are used throughout; this is not full-model GPU residence.

Sources: https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md, https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/configs/ik-llama-baseline-128k.example.sh

## What input was actually completed?

The source reports a 26253-token needle under a 32768 window, a 61-token short smoke under a 131072 window, and a 104805-token exact needle under that longer window. Throughput comes from server timing. Completion lengths, needle depth, per-gate sampling and repeated-run dispersion are absent. A short smoke with a large configured ceiling is not a full-window input test.

Sources: https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md, https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/docs/methodology.md

## Which memory observations are paired with which gate?

Startup, peak RSS, minimum MemAvailable, swap and peak GPU memory are reported for the completed 32K monitoring gate. The longer 104805-token gate has no paired peak telemetry in these selected documents. The MemAvailable sampling boundary is unspecified, so its relationship to the WSL ceiling cannot be assumed. Keep original KiB and MiB labels and do not sum overlapping allocations.

Sources: https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md, https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/docs/limitations.md

## What happened to speculation?

MTP n=4 and canonical n=1 passed the tested correctness checks but failed speed promotion on this particular CPU-expert split. The canonical MTP short gates reached 31775 MiB peak GPU memory. Ngram speculative decoding was not run because its prerequisite failed; it has no measured benefit here. This is separate from the Q8 n-gram embedding table present in the artifact.

Sources: https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md, https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/docs/limitations.md

## Selected source-reported figures

Each point retains its own configuration, measurement unit and source. No site reproduction or cross-source ranking.

```json
[
  {
    "label": "32K needle gate prefill_rate",
    "metric": "prefill_rate",
    "value": 271.64,
    "unit": "tokens/s",
    "statistic": "one reported completed gate; no dispersion",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 26253,
      "context_limit_tokens": 32768,
      "cache_state": null,
      "workload": "32K needle gate",
      "memory_cohort": "completed 32K monitoring"
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 53,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L53"
  },
  {
    "label": "32K needle gate decode_rate",
    "metric": "decode_rate",
    "value": 16.53,
    "unit": "tokens/s",
    "statistic": "one reported completed gate; no dispersion",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 26253,
      "context_limit_tokens": 32768,
      "cache_state": null,
      "workload": "32K needle gate",
      "memory_cohort": "completed 32K monitoring"
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 53,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L53"
  },
  {
    "label": "128K server smoke prefill_rate",
    "metric": "prefill_rate",
    "value": 17.02,
    "unit": "tokens/s",
    "statistic": "one reported completed gate; no dispersion",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 61,
      "context_limit_tokens": 131072,
      "cache_state": null,
      "workload": "128K server smoke",
      "memory_cohort": "no paired 128K telemetry published"
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 54,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L54"
  },
  {
    "label": "128K server smoke decode_rate",
    "metric": "decode_rate",
    "value": 15.26,
    "unit": "tokens/s",
    "statistic": "one reported completed gate; no dispersion",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 61,
      "context_limit_tokens": 131072,
      "cache_state": null,
      "workload": "128K server smoke",
      "memory_cohort": "no paired 128K telemetry published"
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 54,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L54"
  },
  {
    "label": "128K long-context needle prefill_rate",
    "metric": "prefill_rate",
    "value": 245.72,
    "unit": "tokens/s",
    "statistic": "one reported completed gate; no dispersion",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 104805,
      "context_limit_tokens": 131072,
      "cache_state": null,
      "workload": "128K long-context needle",
      "memory_cohort": "no paired 128K telemetry published"
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 55,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L55"
  },
  {
    "label": "128K long-context needle decode_rate",
    "metric": "decode_rate",
    "value": 13.94,
    "unit": "tokens/s",
    "statistic": "one reported completed gate; no dispersion",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 104805,
      "context_limit_tokens": 131072,
      "cache_state": null,
      "workload": "128K long-context needle",
      "memory_cohort": "no paired 128K telemetry published"
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 55,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L55"
  },
  {
    "label": "32K telemetry startup",
    "metric": "startup",
    "value": 25.15,
    "unit": "s",
    "statistic": "source monitored peak/minimum; no distribution",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 26253,
      "context_limit_tokens": 32768,
      "cache_state": null,
      "memory_cohort": "completed 32K gate only; not paired with the 104805-token gate",
      "memory_kind": "startup_duration",
      "MemAvailable_sampling_scope": null,
      "VM_ceiling_and_host_MemAvailable_same_scope_verified": false
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 57,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L57"
  },
  {
    "label": "32K telemetry host_rss",
    "metric": "host_rss",
    "value": 53901956,
    "unit": "KiB",
    "statistic": "source monitored peak/minimum; no distribution",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 26253,
      "context_limit_tokens": 32768,
      "cache_state": null,
      "memory_cohort": "completed 32K gate only; not paired with the 104805-token gate",
      "memory_kind": "peak_host_RSS",
      "MemAvailable_sampling_scope": null,
      "VM_ceiling_and_host_MemAvailable_same_scope_verified": false
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 57,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L57"
  },
  {
    "label": "32K telemetry minimum_memavailable",
    "metric": "minimum_memavailable",
    "value": 86701016,
    "unit": "KiB",
    "statistic": "source monitored peak/minimum; no distribution",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 26253,
      "context_limit_tokens": 32768,
      "cache_state": null,
      "memory_cohort": "completed 32K gate only; not paired with the 104805-token gate",
      "memory_kind": "minimum_MemAvailable_in_reported_monitor_scope",
      "MemAvailable_sampling_scope": null,
      "VM_ceiling_and_host_MemAvailable_same_scope_verified": false
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 57,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L57"
  },
  {
    "label": "32K telemetry swap_use",
    "metric": "swap_use",
    "value": 63708,
    "unit": "KiB",
    "statistic": "source monitored peak/minimum; no distribution",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 26253,
      "context_limit_tokens": 32768,
      "cache_state": null,
      "memory_cohort": "completed 32K gate only; not paired with the 104805-token gate",
      "memory_kind": "maximum_swap_use",
      "MemAvailable_sampling_scope": null,
      "VM_ceiling_and_host_MemAvailable_same_scope_verified": false
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 57,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L57"
  },
  {
    "label": "32K telemetry GPU_memory",
    "metric": "GPU_memory",
    "value": 25813,
    "unit": "MiB",
    "statistic": "source monitored peak/minimum; no distribution",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 0,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": 26253,
      "context_limit_tokens": 32768,
      "cache_state": null,
      "memory_cohort": "completed 32K gate only; not paired with the 104805-token gate",
      "memory_kind": "peak_GPU_memory",
      "MemAvailable_sampling_scope": null,
      "VM_ceiling_and_host_MemAvailable_same_scope_verified": false
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 57,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L57"
  },
  {
    "label": "canonical MTP n1 peak GPU memory",
    "metric": "GPU_memory",
    "value": 31775,
    "unit": "MiB",
    "statistic": "maximum in completed MTP short gates",
    "uncertainty": null,
    "conditions": {
      "model": "Qwen3.8-Flash-Next",
      "model_artifact": "jamesrogers/Qwen3.8-Flash-Next-MTP-MXFP4-GGUF",
      "model_revision": "4a768bf4458a83e1c80667b4d818ec721080546e",
      "artifact_SHA256": "5a4236a6eeda73d23c151f8ed19e5457b0bf1935e1c02c1d42d98a0e26dadde2",
      "engine": "ik_llama.cpp",
      "runtime_execution_revision": "d854dfd36a7262fc46cb7542716b04a9f6d4b47d",
      "MTP_patch_head": "108dfaa77fdaa138c46b043bfff0e94fbe2adff7",
      "system": "single RTX 5090 32607 MiB; 16-core/32-thread desktop CPU; 128 GiB-class host; Ubuntu WSL2 with 96 GiB VM ceiling",
      "node_count": 1,
      "concurrency": 1,
      "configured_max_requests": 1,
      "weight_precision": "MXFP4 GGUF, Q8 n-gram table",
      "CPU_expert_layers": 38,
      "defer_experts": true,
      "mmap": true,
      "GGML_CUDA_NO_PINNED": true,
      "kv_dtype": "Q8_0",
      "draft_tokens": 1,
      "ngram_speculation": false,
      "ngram_embedding_table_present": true,
      "temperature": null,
      "top_p": null,
      "thinking": null,
      "output_tokens": null,
      "actual_output_tokens": null,
      "repetitions": null,
      "experiment_date": null,
      "throughput_origin": "server timing fields, not a client estimate",
      "input_tokens": null,
      "context_limit_tokens": 131072,
      "cache_state": null,
      "memory_cohort": "canonical MTP n1 short gates; not baseline or long-context peak",
      "memory_kind": "peak_GPU_memory"
    },
    "evidence_state": "source_reported",
    "site_reproduced": false,
    "source_document_id": "qwen5090-readme",
    "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
    "source_line": 61,
    "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L61"
  }
]
```

## Supplemental source identities

```json
{
  "supplemental_sources": [
    {
      "document_id": "qwen5090-method",
      "parent_record_id": "qwen38-flash-next-5090-offload-gates",
      "publisher": "kutaelee",
      "repository": "kutaelee/qwen38-flash-next-5090-128k-runtime-recipe",
      "revision": "fca75dd108ce04b6a79db3b45934ed58bba893bb",
      "document_path": "docs/methodology.md",
      "pinned_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/docs/methodology.md",
      "document_sha256": "54d61a37de6873bcebe34c2ad34978a9b0a7191232fb865c364540cdd3e4c706",
      "checked_on": "2026-10-18"
    },
    {
      "document_id": "qwen5090-limits",
      "parent_record_id": "qwen38-flash-next-5090-offload-gates",
      "publisher": "kutaelee",
      "repository": "kutaelee/qwen38-flash-next-5090-128k-runtime-recipe",
      "revision": "fca75dd108ce04b6a79db3b45934ed58bba893bb",
      "document_path": "docs/limitations.md",
      "pinned_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/docs/limitations.md",
      "document_sha256": "029f3cfc9a4b8e2e03eba92137e4df3ca722782b13af4dd990295a8fe1c270d2",
      "checked_on": "2026-10-18"
    },
    {
      "document_id": "qwen5090-config",
      "parent_record_id": "qwen38-flash-next-5090-offload-gates",
      "publisher": "kutaelee",
      "repository": "kutaelee/qwen38-flash-next-5090-128k-runtime-recipe",
      "revision": "fca75dd108ce04b6a79db3b45934ed58bba893bb",
      "document_path": "configs/ik-llama-baseline-128k.example.sh",
      "pinned_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/configs/ik-llama-baseline-128k.example.sh",
      "document_sha256": "fc5de996c4e03b80de59f21597ae543a37ff829822641bf2117426b9c6f09f28",
      "checked_on": "2026-10-18"
    }
  ]
}
```

## memory observations

```json
{
  "memory_observations": [
    {
      "kind": "unpaired_memory_telemetry",
      "measured_context_limit_tokens": 32768,
      "measured_input_tokens": 26253,
      "longer_needle_input_tokens": 104805,
      "longer_needle_peak_RAM": null,
      "host_MemAvailable_scope": null,
      "CPU_expert_layers": 38,
      "source_GGUF_bytes": 126846153856
    }
  ]
}
```

## long context cases

```json
{
  "long_context_cases": [
    {
      "classification": "source_reported_exact_needle_success",
      "actual_input_tokens": 26253,
      "configured_limit_tokens": 32768,
      "actual_output_tokens": null,
      "exact_match": true,
      "needle_depth_fraction": null,
      "general_reasoning_quality_proven": false,
      "source_document_id": "qwen5090-readme",
      "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
      "source_line": 53,
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L53"
    },
    {
      "classification": "source_reported_short_smoke",
      "actual_input_tokens": 61,
      "configured_limit_tokens": 131072,
      "actual_output_tokens": null,
      "exact_match": true,
      "needle_depth_fraction": null,
      "general_reasoning_quality_proven": false,
      "source_document_id": "qwen5090-readme",
      "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
      "source_line": 54,
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L54"
    },
    {
      "classification": "source_reported_exact_needle_success",
      "actual_input_tokens": 104805,
      "configured_limit_tokens": 131072,
      "actual_output_tokens": null,
      "exact_match": true,
      "needle_depth_fraction": null,
      "general_reasoning_quality_proven": false,
      "source_document_id": "qwen5090-readme",
      "source_document_sha256": "0668bd3d2e2ca6ed48b4ef13e9d9c56e4ce5c37c45ebbf8dc54e359bd9575de0",
      "source_line": 55,
      "evidence_url": "https://github.com/kutaelee/qwen38-flash-next-5090-128k-runtime-recipe/blob/fca75dd108ce04b6a79db3b45934ed58bba893bb/README.md#L55"
    }
  ]
}
```